View
220
Download
4
Tags:
Embed Size (px)
Citation preview
CSE 5331/7331 F'2011 1
CSE 5331/7331Fall 2011
DATA MININGIntroductory and Related Topics
Margaret H. DunhamDepartment of Computer Science and Engineering
Southern Methodist University
Slides extracted from Data Mining, Introductory and Advanced Topics, Prentice Hall, 2002.
2CSE 5331/7331 F'2011
Data Mining Outline
PART I – Introduction– Techniques
PART II – Core Topics PART III – Related Topics
3CSE 5331/7331 F'2011
Introduction Outline
Define data mining Data mining vs. databases Basic data mining tasks Data mining development Data mining issues
Goal: Provide an overview of data mining.
4CSE 5331/7331 F'2011
Introduction
Data is growing at a phenomenal rate Users expect more sophisticated
information How?
UNCOVER HIDDEN INFORMATION
DATA MINING
5CSE 5331/7331 F'2011
Data Mining Definition
Finding hidden information in a database
Fit data to a model Similar terms
– Exploratory data analysis– Data driven discovery– Deductive learning
6CSE 5331/7331 F'2011
Data Mining Algorithm
Objective: Fit Data to a Model– Descriptive– Predictive
Preference – Technique to choose the best model
Search – Technique to search the data– “Query”
7CSE 5331/7331 F'2011
Database Processing vs. Data Mining Processing
Query– Well defined– SQL
Query– Poorly defined– No precise query
language Data– Operational data
Output– Precise– Subset of database
Data– Not operational data
Output– Fuzzy– Not a subset of database
8CSE 5331/7331 F'2011
Query Examples Database
Data Mining
– Find all customers who have purchased milk
– Find all items which are frequently purchased with milk. (association rules)
– Find all credit applicants with last name of Smith.– Identify customers who have purchased more
than $10,000 in the last month.
– Find all credit applicants who are poor credit risks. (classification)
– Identify customers with similar buying habits. (Clustering)
10CSE 5331/7331 F'2011
Basic Data Mining Tasks Classification maps data into predefined
groups or classes– Supervised learning– Pattern recognition– Prediction
Regression is used to map a data item to a real valued prediction variable.
Clustering groups similar data together into clusters.– Unsupervised learning– Segmentation– Partitioning
11CSE 5331/7331 F'2011
Basic Data Mining Tasks (cont’d)
Summarization maps data into subsets with associated simple descriptions.– Characterization– Generalization
Link Analysis uncovers relationships among data.– Affinity Analysis– Association Rules– Sequential Analysis determines sequential
patterns.
12CSE 5331/7331 F'2011
Ex: Time Series Analysis Example: Stock Market Predict future values Determine similar patterns over time Classify behavior
13CSE 5331/7331 F'2011
Data Mining vs. KDD
Knowledge Discovery in Databases (KDD): process of finding useful information and patterns in data.
Data Mining: Use of algorithms to extract the information and patterns derived by the KDD process.
14CSE 5331/7331 F'2011
KDD Process
Selection: Obtain data from various sources. Preprocessing: Cleanse data. Transformation: Convert to common format.
Transform to new format. Data Mining: Obtain desired results. Interpretation/Evaluation: Present results
to user in meaningful manner.
Modified from [FPSS96C]
15CSE 5331/7331 F'2011
KDD Process Ex: Web Log Selection:
– Select log data (dates and locations) to use Preprocessing:
– Remove identifying URLs– Remove error logs
Transformation: – Sessionize (sort and group)
Data Mining: – Identify and count patterns– Construct data structure
Interpretation/Evaluation:– Identify and display frequently accessed sequences.
Potential User Applications:– Cache prediction– Personalization
16CSE 5331/7331 F'2011
Data Mining Development• Similarity Measures• Hierarchical Clustering• IR Systems• Imprecise Queries• Textual Data• Web Search Engines
• Bayes Theorem• Regression Analysis• EM Algorithm• K-Means Clustering• Time Series Analysis
• Neural Networks• Decision Tree Algorithms
• Algorithm Design Techniques• Algorithm Analysis• Data Structures
• Relational Data Model• SQL• Association Rule
Algorithms• Data Warehousing• Scalability Techniques
17CSE 5331/7331 F'2011
KDD Issues
Human Interaction Overfitting Outliers Interpretation Visualization Large Datasets High Dimensionality
18CSE 5331/7331 F'2011
KDD Issues (cont’d)
Multimedia Data Missing Data Irrelevant Data Noisy Data Changing Data Integration Application
20CSE 5331/7331 F'2011
Data Mining Metrics
Usefulness Return on Investment (ROI) Accuracy Space/Time
21CSE 5331/7331 F'2011
Visualization Techniques
Graphical Geometric Icon-based Pixel-based Hierarchical Hybrid
22CSE 5331/7331 F'2011
Models Based on Summarization
Visualization: Frequency distribution, mean, variance, median, mode, etc.
Box Plot:
24CSE 5331/7331 F'2011
Data Mining Techniques Outline
Statistical– Point Estimation– Models Based on Summarization– Bayes Theorem– Hypothesis Testing– Regression and Correlation
Similarity Measures Decision Trees Neural Networks
– Activation Functions Genetic Algorithms
Goal: Provide an overview of basic data mining techniques
25CSE 5331/7331 F'2011
Point Estimation Point Estimate: estimate a population
parameter. May be made by calculating the parameter for a
sample. May be used to predict value for missing data. Ex:
– R contains 100 employees– 99 have salary information– Mean salary of these is $50,000– Use $50,000 as value of remaining employee’s
salary. Is this a good idea?
26CSE 5331/7331 F'2011
Estimation Error
Bias: Difference between expected value and actual value.
Mean Squared Error (MSE): expected value of the squared difference between the estimate and the actual value:
Why square? Root Mean Square Error (RMSE)
27CSE 5331/7331 F'2011
Jackknife Estimate Jackknife Estimate: estimate of parameter is
obtained by omitting one value from the set of observed values.
Ex: estimate of mean for X={x1, … , xn}
28CSE 5331/7331 F'2011
Maximum Likelihood Estimate (MLE)
Obtain parameter estimates that maximize the probability that the sample data occurs for the specific model.
Joint probability for observing the sample data by multiplying the individual probabilities. Likelihood function:
Maximize L.
29CSE 5331/7331 F'2011
MLE Example
Coin toss five times: {H,H,H,H,T}
Assuming a perfect coin with H and T equally
likely, the likelihood of this sequence is:
However if the probability of a H is 0.8 then:
30CSE 5331/7331 F'2011
MLE Example (cont’d) General likelihood formula:
Estimate for p is then 4/5 = 0.8
31CSE 5331/7331 F'2011
Expectation-Maximization (EM)
Solves estimation with incomplete data. Obtain initial estimates for parameters. Iteratively use estimates for missing
data and continue until convergence.
34CSE 5331/7331 F'2011
Bayes Theorem
Posterior Probability: P(h1|xi) Prior Probability: P(h1) Bayes Theorem:
Assign probabilities of hypotheses given a data value.
35CSE 5331/7331 F'2011
Bayes Theorem Example Credit authorizations (hypotheses):
h1=authorize purchase, h2 = authorize after further identification, h3=do not authorize, h4= do not authorize but contact police
Assign twelve data values for all combinations of credit and income:
From training data: P(h1) = 60%; P(h2)=20%;
P(h3)=10%; P(h4)=10%.
1 2 3 4 Excellent x1 x2 x3 x4 Good x5 x6 x7 x8 Bad x9 x10 x11 x12
36CSE 5331/7331 F'2011
Bayes Example(cont’d) Training Data:
ID Income Credit Class xi 1 4 Excellent h1 x4 2 3 Good h1 x7 3 2 Excellent h1 x2 4 3 Good h1 x7 5 4 Good h1 x8 6 2 Excellent h1 x2 7 3 Bad h2 x11 8 2 Bad h2 x10 9 3 Bad h3 x11 10 1 Bad h4 x9
37CSE 5331/7331 F'2011
Bayes Example(cont’d) Calculate P(xi|hj) and P(xi)
Ex: P(x7|h1)=2/6; P(x4|h1)=1/6; P(x2|h1)=2/6; P(x8|h1)=1/6; P(xi|h1)=0 for all other xi.
Predict the class for x4:– Calculate P(hj|x4) for all hj. – Place x4 in class with largest value.– Ex:
»P(h1|x4)=(P(x4|h1)(P(h1))/P(x4) =(1/6)(0.6)/0.1=1.
»x4 in class h1.
38CSE 5331/7331 F'2011
Regression
Predict future values based on past values
Linear Regression assumes linear relationship exists.
y = c0 + c1 x1 + … + cn xn
Find values to best fit the data
40CSE 5331/7331 F'2011
Correlation
Examine the degree to which the values for two variables behave similarly.
Correlation coefficient r:• 1 = perfect correlation• -1 = perfect but opposite correlation• 0 = no correlation
41CSE 5331/7331 F'2011
Similarity Measures
Determine similarity between two objects. Similarity characteristics:
Alternatively, distance measure measure how unlike or dissimilar objects are.
45CSE 5331/7331 F'2011
Decision Trees Decision Tree (DT):
– Tree where the root and each internal node is labeled with a question.
– The arcs represent each possible answer to the associated question.
– Each leaf node represents a prediction of a solution to the problem.
Popular technique for classification; Leaf node indicates class to which the corresponding tuple belongs.
47CSE 5331/7331 F'2011
Decision Trees
A Decision Tree Model is a computational model consisting of three parts:– Decision Tree– Algorithm to create the tree– Algorithm that applies the tree to data
Creation of the tree is the most difficult part. Processing is basically a search similar to
that in a binary search tree (although DT may not be binary).
49CSE 5331/7331 F'2011
DT Advantages/Disadvantages
Advantages:– Easy to understand. – Easy to generate rules
Disadvantages:– May suffer from overfitting.– Classifies by rectangular partitioning.– Does not easily handle nonnumeric data.– Can be quite large – pruning is necessary.
50CSE 5331/7331 F'2011
Neural Networks Based on observed functioning of human
brain. (Artificial Neural Networks (ANN) Our view of neural networks is very
simplistic. We view a neural network (NN) from a
graphical viewpoint. Alternatively, a NN may be viewed from
the perspective of matrices. Used in pattern recognition, speech
recognition, computer vision, and classification.
51CSE 5331/7331 F'2011
Neural Networks Neural Network (NN) is a directed graph
F=<V,A> with vertices V={1,2,…,n} and arcs A={<i,j>|1<=i,j<=n}, with the following restrictions:– V is partitioned into a set of input nodes, VI,
hidden nodes, VH, and output nodes, VO.– The vertices are also partitioned into layers – Any arc <i,j> must have node i in layer h-1
and node j in layer h.– Arc <i,j> is labeled with a numeric value wij.– Node i is labeled with a function fi.
54CSE 5331/7331 F'2011
NN Activation Functions
Functions associated with nodes in graph.
Output may be in range [-1,1] or [0,1]
56CSE 5331/7331 F'2011
NN Learning
Propagate input values through graph. Compare output to desired output. Adjust weights in graph accordingly.
57CSE 5331/7331 F'2011
Neural Networks
A Neural Network Model is a computational model consisting of three parts:– Neural Network graph – Learning algorithm that indicates how
learning takes place.– Recall techniques that determine hew
information is obtained from the network. We will look at propagation as the recall
technique.
58CSE 5331/7331 F'2011
NN Advantages
Learning Can continue learning even after
training set has been applied. Easy parallelization Solves many problems
59CSE 5331/7331 F'2011
NN Disadvantages
Difficult to understand May suffer from overfitting Structure of graph must be determined
a priori. Input values must be numeric. Verification difficult.
60CSE 5331/7331 F'2011
Genetic Algorithms Optimization search type algorithms. Creates an initial feasible solution and
iteratively creates new “better” solutions. Based on human evolution and survival of the
fittest. Must represent a solution as an individual. Individual: string I=I1,I2,…,In where Ij is in
given alphabet A. Each character Ij is called a gene. Population: set of individuals.
61CSE 5331/7331 F'2011
Genetic Algorithms A Genetic Algorithm (GA) is a computational
model consisting of five parts:– A starting set of individuals, P.– Crossover: technique to combine two
parents to create offspring.– Mutation: randomly change an individual.– Fitness: determine the best individuals.– Algorithm which applies the crossover and
mutation techniques to P iteratively using the fitness function to determine the best individuals in P to keep.
62CSE 5331/7331 F'2011
Crossover Examples
111 111
000 000
Parents Children
111 000
000 111
a) Single Crossover
111 111
Parents Children
111 000
000
a) Single Crossover
111 111
000 000
Parents
a) Multiple Crossover
111 111
000
Parents Children
111 000
000 111
Children
111 000
000 11100
11
00
11
64CSE 5331/7331 F'2011
GA Advantages/Disadvantages Advantages
– Easily parallelized Disadvantages
– Difficult to understand and explain to end users.
– Abstraction of the problem and method to represent individuals is quite difficult.
– Determining fitness function is difficult.– Determining how to perform crossover and
mutation is difficult.
65CSE 5331/7331 F'2011
Data Mining Outline
PART I - Introduction PART II – Core Topics
– Classification– Clustering– Association Rules
PART III – Related Topics
66CSE 5331/7331 F'2011
Classification Outline
Classification Problem Overview Classification Techniques
– Regression– Distance– Decision Trees– Rules– Neural Networks
Goal: Provide an overview of the classification problem and introduce some of the basic algorithms
67CSE 5331/7331 F'2011
Classification Problem Given a database D={t1,t2,…,tn} and a set
of classes C={C1,…,Cm}, the Classification Problem is to define a mapping f:DgC where each ti is assigned to one class.
Actually divides D into equivalence classes.
Prediction is similar, but may be viewed as having infinite number of classes.
68CSE 5331/7331 F'2011
Classification Examples
Teachers classify students’ grades as A, B, C, D, or F.
Identify mushrooms as poisonous or edible.
Predict when a river will flood. Identify individuals with credit risks. Speech recognition Pattern recognition
69CSE 5331/7331 F'2011
Classification Ex: Grading
If x >= 90 then grade =A.
If 80<=x<90 then grade =B.
If 70<=x<80 then grade =C.
If 60<=x<70 then grade =D.
If x<50 then grade =F.
>=90<90
x
>=80<80
x
>=70<70
x
F
B
A
>=60<50
x C
D
70CSE 5331/7331 F'2011
Classification Ex: Letter Recognition
View letters as constructed from 5 components:
Letter C
Letter E
Letter A
Letter D
Letter F
Letter B
71CSE 5331/7331 F'2011
Classification Techniques
Approach:1. Create specific model by evaluating
training data (or using domain experts’ knowledge).
2. Apply model developed to new data. Classes must be predefined Most common techniques use DTs,
NNs, or are based on distances or statistical methods.
73CSE 5331/7331 F'2011
Issues in Classification
Missing Data– Ignore– Replace with assumed value
Measuring Performance– Classification accuracy on test data– Confusion matrix– OC Curve
74CSE 5331/7331 F'2011
Height Example DataName Gender Height Output1 Output2 Kristina F 1.6m Short Medium Jim M 2m Tall Medium Maggie F 1.9m Medium Tall Martha F 1.88m Medium Tall Stephanie F 1.7m Short Medium Bob M 1.85m Medium Medium Kathy F 1.6m Short Medium Dave M 1.7m Short Medium Worth M 2.2m Tall Tall Steven M 2.1m Tall Tall Debbie F 1.8m Medium Medium Todd M 1.95m Medium Medium Kim F 1.9m Medium Tall Amy F 1.8m Medium Medium Wynette F 1.75m Medium Medium
75CSE 5331/7331 F'2011
Classification Performance
True Positive
True NegativeFalse Positive
False Negative
76CSE 5331/7331 F'2011
Confusion Matrix Example
Using height data example with Output1 correct and Output2 actual assignment
Actual Assignment Membership Short Medium Tall Short 0 4 0 Medium 0 5 3 Tall 0 1 2
78CSE 5331/7331 F'2011
RegressionTopics
Linear Regression Nonlinear Regression Logistic Regression Metrics
79CSE 5331/7331 F'2011
Remember High School?
Y= mx + b You need two points to determine a
straight line. You need two points to find values for m
and b.
THIS IS REGRESSION
80CSE 5331/7331 F'2011
Regression Assume data fits a predefined function Determine best values for regression
coefficients c0,c1,…,cn. Assume an error: y = c0+c1x1+…+cnxn+e Estimate error using mean squared error for
training set:
81CSE 5331/7331 F'2011
Linear Regression Assume data fits a predefined function Determine best values for regression
coefficients c0,c1,…,cn. Assume an error: y = c0+c1x1+…+cnxn+e Estimate error using mean squared error for
training set:
82CSE 5331/7331 F'2011
Classification Using Linear Regression
Division: Use regression function to divide area into regions.
Prediction: Use regression function to predict a class membership function. Input includes desired class.
85CSE 5331/7331 F'2011
Linear Regression Poor Fit
Why use sum of least squares?http://curvefit.com/sum_of_squares.htmLinear doesn’t always work well
86CSE 5331/7331 F'2011
Nonlinear Regression
Data does not nicely fit a straight line Fit data to a curve Many possible functions Not as easy and straightforward as
linear regression How nonlinear regression works:
http://curvefit.com/how_nonlin_works.htm
87CSE 5331/7331 F'2011
P-value
The probability that a variable has a value greater than the observed value
http://en.wikipedia.org/wiki/P-value http://sportsci.org/resource/stats/pvalues.html
88CSE 5331/7331 F'2011
Covariance
Degree to which two variables vary in the same manner
Correlation is normalized and covariance is not
http://www.ds.unifi.it/VL/VL_EN/expect/expect3.html
89CSE 5331/7331 F'2011
Residual
Error Difference between desired output and
predicted output May actually use sum of squares
90CSE 5331/7331 F'2011
Classification Using Distance Place items in class to which they are
“closest”. Must determine distance between an
item and a class. Classes represented by
–Centroid: Central value.–Medoid: Representative point.– Individual points
Algorithm: KNN
91CSE 5331/7331 F'2011
K Nearest Neighbor (KNN):
Training set includes classes. Examine K items near item to be
classified. New item placed in class with the most
number of close items. O(q) for each tuple to be classified.
(Here q is the size of the training set.)
94CSE 5331/7331 F'2011
Classification Using Decision Trees
Partitioning based: Divide search space into rectangular regions.
Tuple placed into class based on the region within which it falls.
DT approaches differ in how the tree is built: DT Induction
Internal nodes associated with attribute and arcs with values for that attribute.
Algorithms: ID3, C4.5, CART
95CSE 5331/7331 F'2011
Decision TreeGiven:
– D = {t1, …, tn} where ti=<ti1, …, tih> – Database schema contains {A1, A2, …, Ah}– Classes C={C1, …., Cm}
Decision or Classification Tree is a tree associated with D such that– Each internal node is labeled with attribute, Ai
– Each arc is labeled with predicate which can be applied to attribute at parent
– Each leaf node is labeled with a class, Cj
99CSE 5331/7331 F'2011
DT Issues
Choosing Splitting Attributes Ordering of Splitting Attributes Splits Tree Structure Stopping Criteria Training Data Pruning
102CSE 5331/7331 F'2011
DT Induction
When all the marbles in the bowl are mixed up, little information is given.
When the marbles in the bowl are all from one class and those in the other two classes are on either side, more information is given.
Use this approach with DT Induction !
103CSE 5331/7331 F'2011
Information/Entropy Given probabilitites p1, p2, .., ps whose sum is
1, Entropy is defined as:
Entropy measures the amount of randomness or surprise or uncertainty.
Goal in classification– no surprise– entropy = 0
105CSE 5331/7331 F'2011
ID3 Creates tree using information theory
concepts and tries to reduce expected number of comparison..
ID3 chooses split attribute with the highest information gain:
106CSE 5331/7331 F'2011
ID3 Example (Output1) Starting state entropy:4/15 log(15/4) + 8/15 log(15/8) + 3/15 log(15/3) = 0.4384 Gain using gender:
– Female: 3/9 log(9/3)+6/9 log(9/6)=0.2764– Male: 1/6 (log 6/1) + 2/6 log(6/2) + 3/6 log(6/3) =
0.4392– Weighted sum: (9/15)(0.2764) + (6/15)(0.4392) =
0.34152– Gain: 0.4384 – 0.34152 = 0.09688
Gain using height:0.4384 – (2/15)(0.301) = 0.3983
Choose height as first splitting attribute
107CSE 5331/7331 F'2011
C4.5 ID3 favors attributes with large number of
divisions Improved version of ID3:
– Missing Data– Continuous Data– Pruning– Rules– GainRatio:
108CSE 5331/7331 F'2011
CART
Create Binary Tree Uses entropy Formula to choose split point, s, for node t:
PL,PR probability that a tuple in the training set will be on the left or right side of the tree.
109CSE 5331/7331 F'2011
CART Example At the start, there are six choices for
split point (right branch on equality):– P(Gender)=2(6/15)(9/15)(2/15 + 4/15 + 3/15)=0.224– P(1.6) = 0– P(1.7) = 2(2/15)(13/15)(0 + 8/15 + 3/15) = 0.169– P(1.8) = 2(5/15)(10/15)(4/15 + 6/15 + 3/15) = 0.385– P(1.9) = 2(9/15)(6/15)(4/15 + 2/15 + 3/15) = 0.256– P(2.0) = 2(12/15)(3/15)(4/15 + 8/15 + 3/15) = 0.32
Split at 1.8
110CSE 5331/7331 F'2011
Classification Using Neural Networks
Typical NN structure for classification:– One output node per class– Output value is class membership function value
Supervised learning For each tuple in training set, propagate it
through NN. Adjust weights on edges to improve future classification.
Algorithms: Propagation, Backpropagation, Gradient Descent
111CSE 5331/7331 F'2011
NN Issues
Number of source nodes Number of hidden layers Training data Number of sinks Interconnections Weights Activation Functions Learning Technique When to stop learning
116CSE 5331/7331 F'2011
NN Learning
Adjust weights to perform better with the associated test data.
Supervised: Use feedback from knowledge of correct classification.
Unsupervised: No knowledge of correct classification needed.
118CSE 5331/7331 F'2011
Supervised Learning
Possible error values assuming output from node i is yi but should be di:
Change weights on arcs based on estimated error
119CSE 5331/7331 F'2011
NN Backpropagation
Propagate changes to weights backward from output layer to input layer.
Delta Rule: r wij= c xij (dj – yj) Gradient Descent: technique to modify
the weights in the graph.
126CSE 5331/7331 F'2011
Types of NNs
Different NN structures used for different problems.
Perceptron Self Organizing Feature Map Radial Basis Function Network
128CSE 5331/7331 F'2011
Perceptron Example
Suppose:– Summation: S=3x1+2x2-6
– Activation: if S>0 then 1 else 0
129CSE 5331/7331 F'2011
Self Organizing Feature Map (SOFM)
Competitive Unsupervised Learning Observe how neurons work in brain:
– Firing impacts firing of those near– Neurons far apart inhibit each other– Neurons have specific nonoverlapping
tasks Ex: Kohonen Network
131CSE 5331/7331 F'2011
Kohonen Network
Competitive Layer – viewed as 2D grid Similarity between competitive nodes and
input nodes:– Input: X = <x1, …, xh>
– Weights: <w1i, … , whi>
– Similarity defined based on dot product Competitive node most similar to input “wins” Winning node weights (as well as
surrounding node weights) increased.
132CSE 5331/7331 F'2011
Radial Basis Function Network
RBF function has Gaussian shape RBF Networks
– Three Layers– Hidden layer – Gaussian activation
function– Output layer – Linear activation function
134CSE 5331/7331 F'2011
Classification Using Rules Perform classification using If-Then
rules Classification Rule: r = <a,c>
Antecedent, Consequent May generate from from other
techniques (DT, NN) or generate directly.
Algorithms: Gen, RX, 1R, PRISM
142CSE 5331/7331 F'2011
Decision Tree vs. Rules
Tree has implied order in which splitting is performed.
Tree created based on looking at all classes.
Rules have no ordering of predicates.
Only need to look at one class to generate its rules.
143CSE 5331/7331 F'2011
Clustering Outline
Clustering Problem Overview Clustering Techniques
– Hierarchical Algorithms– Partitional Algorithms– Genetic Algorithm– Clustering Large Databases
Goal: Provide an overview of the clustering problem and introduce some of the basic algorithms
144CSE 5331/7331 F'2011
Clustering Examples
Segment customer database based on similar buying patterns.
Group houses in a town into neighborhoods based on similar features.
Identify new plant species Identify similar Web usage patterns
147CSE 5331/7331 F'2011
Clustering vs. Classification
No prior knowledge– Number of clusters– Meaning of clusters
Unsupervised learning
148CSE 5331/7331 F'2011
Clustering Issues
Outlier handling Dynamic data Interpreting results Evaluating results Number of clusters Data to be used Scalability
150CSE 5331/7331 F'2011
Clustering Problem
Given a database D={t1,t2,…,tn} of tuples and an integer value k, the Clustering Problem is to define a mapping f:Dg{1,..,k} where each ti is assigned to one cluster Kj, 1<=j<=k.
A Cluster, Kj, contains precisely those tuples mapped to it.
Unlike classification problem, clusters are not known a priori.
151CSE 5331/7331 F'2011
Types of Clustering
Hierarchical – Nested set of clusters created.
Partitional – One set of clusters created.
Incremental – Each element handled one at a time.
Simultaneous – All elements handled together.
Overlapping/Non-overlapping
153CSE 5331/7331 F'2011
Distance Between Clusters Single Link: smallest distance between
points Complete Link: largest distance between
points Average Link: average distance between
points Centroid: distance between centroids
154CSE 5331/7331 F'2011
Hierarchical Clustering
Clusters are created in levels actually creating sets of clusters at each level.
Agglomerative– Initially each item in its own cluster– Iteratively clusters are merged together– Bottom Up
Divisive– Initially all items in one cluster– Large clusters are successively divided– Top Down
155CSE 5331/7331 F'2011
Hierarchical Algorithms
Single Link MST Single Link Complete Link Average Link
156CSE 5331/7331 F'2011
Dendrogram
Dendrogram: a tree data structure which illustrates hierarchical clustering techniques.
Each level shows clusters for that level.– Leaf – individual clusters– Root – one cluster
A cluster at level i is the union of its children clusters at level i+1.
158CSE 5331/7331 F'2011
Agglomerative ExampleA B C D E
A 0 1 2 2 3
B 1 0 2 4 3
C 2 2 0 1 5
D 2 4 1 0 3
E 3 3 5 3 0
BA
E C
D
4
Threshold of
2 3 51
A B C D E
159CSE 5331/7331 F'2011
MST Example
A B C D E
A 0 1 2 2 3
B 1 0 2 4 3
C 2 2 0 1 5
D 2 4 1 0 3
E 3 3 5 3 0
BA
E C
D
161CSE 5331/7331 F'2011
Single Link View all items with links (distances)
between them. Finds maximal connected components
in this graph. Two clusters are merged if there is at
least one edge which connects them. Uses threshold distances at each level. Could be agglomerative or divisive.
164CSE 5331/7331 F'2011
Partitional Clustering
Nonhierarchical Creates clusters in one step as
opposed to several steps. Since only one set of clusters is output,
the user normally has to input the desired number of clusters, k.
Usually deals with static sets.
165CSE 5331/7331 F'2011
Partitional Algorithms
MST Squared Error K-Means Nearest Neighbor PAM BEA GA
169CSE 5331/7331 F'2011
K-Means Initial set of clusters randomly chosen. Iteratively, items are moved among sets
of clusters until the desired set is reached.
High degree of similarity among elements in a cluster is obtained.
Given a cluster Ki={ti1,ti2,…,tim}, the
cluster mean is mi = (1/m)(ti1 + … + tim)
170CSE 5331/7331 F'2011
K-Means Example Given: {2,4,10,12,3,20,30,11,25}, k=2 Randomly assign means: m1=3,m2=4 K1={2,3}, K2={4,10,12,20,30,11,25},
m1=2.5,m2=16 K1={2,3,4},K2={10,12,20,30,11,25},
m1=3,m2=18 K1={2,3,4,10},K2={12,20,30,11,25},
m1=4.75,m2=19.6 K1={2,3,4,10,11,12},K2={20,30,25},
m1=7,m2=25 Stop as the clusters with these means
are the same.
172CSE 5331/7331 F'2011
Nearest Neighbor
Items are iteratively merged into the existing clusters that are closest.
Incremental Threshold, t, used to determine if items
are added to existing clusters or a new cluster is created.
174CSE 5331/7331 F'2011
PAM
Partitioning Around Medoids (PAM) (K-Medoids)
Handles outliers well. Ordering of input does not impact results. Does not scale well. Each cluster represented by one item,
called the medoid. Initial set of k medoids randomly chosen.
176CSE 5331/7331 F'2011
PAM Cost Calculation At each step in algorithm, medoids are
changed if the overall cost is improved. Cjih – cost change for an item tj associated
with swapping medoid ti with non-medoid th.
178CSE 5331/7331 F'2011
BEA Bond Energy Algorithm Database design (physical and logical) Vertical fragmentation Determine affinity (bond) between attributes
based on common usage. Algorithm outline:
1. Create affinity matrix
2. Convert to BOND matrix
3. Create regions of close bonding
180CSE 5331/7331 F'2011
Genetic Algorithm Example
{A,B,C,D,E,F,G,H} Randomly choose initial solution:
{A,C,E} {B,F} {D,G,H} or10101000, 01000100, 00010011
Suppose crossover at point four and choose 1st and 3rd individuals:10100011, 01000100, 00011000
What should termination criteria be?
182CSE 5331/7331 F'2011
Clustering Large Databases
Most clustering algorithms assume a large data structure which is memory resident.
Clustering may be performed first on a sample of the database then applied to the entire database.
Algorithms– BIRCH– DBSCAN– CURE
183CSE 5331/7331 F'2011
Desired Features for Large Databases
One scan (or less) of DB Online Suspendable, stoppable, resumable Incremental Work with limited main memory Different techniques to scan (e.g.
sampling) Process each tuple once
184CSE 5331/7331 F'2011
BIRCH Balanced Iterative Reducing and
Clustering using Hierarchies Incremental, hierarchical, one scan Save clustering information in a tree Each entry in the tree contains
information about one cluster New nodes inserted in closest entry in
tree
185CSE 5331/7331 F'2011
Clustering Feature CT Triple: (N,LS,SS)
– N: Number of points in cluster– LS: Sum of points in the cluster– SS: Sum of squares of points in the cluster
CF Tree– Balanced search tree– Node has CF triple for each child– Leaf node represents cluster and has CF value
for each subcluster in it.– Subcluster has maximum diameter
188CSE 5331/7331 F'2011
DBSCAN
Density Based Spatial Clustering of Applications with Noise
Outliers will not effect creation of cluster. Input
– MinPts – minimum number of points in cluster
– Eps – for each point in cluster there must be another point in it less than this distance away.
189CSE 5331/7331 F'2011
DBSCAN Density Concepts
Eps-neighborhood: Points within Eps distance of a point.
Core point: Eps-neighborhood dense enough (MinPts)
Directly density-reachable: A point p is directly density-reachable from a point q if the distance is small (Eps) and q is a core point.
Density-reachable: A point si density-reachable form another point if there is a path from one to the other consisting of only core points.
192CSE 5331/7331 F'2011
CURE
Clustering Using Representatives Use many points to represent a cluster
instead of only one Points will be well scattered
197CSE 5331/7331 F'2011
Association Rules OutlineGoal: Provide an overview of basic
Association Rule mining techniques Association Rules Problem Overview
– Large itemsets Association Rules Algorithms
– Apriori– Sampling– Partitioning– Parallel Algorithms
Comparing Techniques Incremental Algorithms Advanced AR Techniques
198CSE 5331/7331 F'2011
Example: Market Basket Data Items frequently purchased together:
Bread PeanutButter Uses:
– Placement – Advertising– Sales– Coupons
Objective: increase sales and reduce costs
199CSE 5331/7331 F'2011
Association Rule Definitions
Set of items: I={I1,I2,…,Im}
Transactions: D={t1,t2, …, tn}, tj I Itemset: {Ii1,Ii2, …, Iik} I Support of an itemset: Percentage of
transactions which contain that itemset. Large (Frequent) itemset: Itemset
whose number of occurrences is above a threshold.
200CSE 5331/7331 F'2011
Association Rules Example
I = { Beer, Bread, Jelly, Milk, PeanutButter}
Support of {Bread,PeanutButter} is 60%
201CSE 5331/7331 F'2011
Association Rule Definitions
Association Rule (AR): implication X Y where X,Y I and X Y = ;
Support of AR (s) X Y: Percentage of transactions that contain X Y
Confidence of AR (a) X Y: Ratio of number of transactions that contain X Y to the number that contain X
203CSE 5331/7331 F'2011
Association Rule Problem Given a set of items I={I1,I2,…,Im} and a
database of transactions D={t1,t2, …, tn} where ti={Ii1,Ii2, …, Iik} and Iij I, the Association Rule Problem is to identify all association rules X Y with a minimum support and confidence.
Link Analysis NOTE: Support of X Y is same as
support of X Y.
204CSE 5331/7331 F'2011
Association Rule Techniques
1. Find Large Itemsets.
2. Generate rules from frequent itemsets.
206CSE 5331/7331 F'2011
Apriori
Large Itemset Property:
Any subset of a large itemset is large. Contrapositive:
If an itemset is not large,
none of its supersets are large.
209CSE 5331/7331 F'2011
Apriori Algorithm
1. C1 = Itemsets of size one in I;
2. Determine all large itemsets of size 1, L1;
3. i = 1;
4. Repeat
5. i = i + 1;
6. Ci = Apriori-Gen(Li-1);
7. Count Ci to determine Li;
8. until no more large itemsets found;
210CSE 5331/7331 F'2011
Apriori-Gen
Generate candidates of size i+1 from large itemsets of size i.
Approach used: join large itemsets of size i if they agree on i-1
May also prune candidates who have subsets that are not large.
213CSE 5331/7331 F'2011
Apriori Adv/Disadv
Advantages:– Uses large itemset property.– Easily parallelized– Easy to implement.
Disadvantages:– Assumes transaction database is memory
resident.– Requires up to m database scans.
214CSE 5331/7331 F'2011
Sampling Large databases Sample the database and apply Apriori to the
sample. Potentially Large Itemsets (PL): Large
itemsets from sample Negative Border (BD - ):
– Generalization of Apriori-Gen applied to itemsets of varying sizes.
– Minimal set of itemsets which are not in PL, but whose subsets are all in PL.
216CSE 5331/7331 F'2011
Sampling Algorithm
1. Ds = sample of Database D;
2. PL = Large itemsets in Ds using smalls;
3. C = PL BD-(PL);4. Count C in Database using s;
5. ML = large itemsets in BD-(PL);6. If ML = then done7. else C = repeated application of
BD-;
8. Count C in Database;
217CSE 5331/7331 F'2011
Sampling Example Find AR assuming s = 20% Ds = { t1,t2} Smalls = 10% PL = {{Bread}, {Jelly}, {PeanutButter},
{Bread,Jelly}, {Bread,PeanutButter}, {Jelly, PeanutButter}, {Bread,Jelly,PeanutButter}}
BD-(PL)={{Beer},{Milk}} ML = {{Beer}, {Milk}} Repeated application of BD- generates all
remaining itemsets
218CSE 5331/7331 F'2011
Sampling Adv/Disadv
Advantages:– Reduces number of database scans to one
in the best case and two in worst.– Scales better.
Disadvantages:– Potentially large number of candidates in
second pass
219CSE 5331/7331 F'2011
Partitioning
Divide database into partitions D1,D2,…,Dp
Apply Apriori to each partition Any large itemset must be large in at
least one partition.
220CSE 5331/7331 F'2011
Partitioning Algorithm
1. Divide D into partitions D1,D2,…,Dp;
2. For I = 1 to p do
3. Li = Apriori(Di);
4. C = L1 … Lp;
5. Count C on D to generate L;
221CSE 5331/7331 F'2011
Partitioning Example
D1
D2
S=10%
L1 ={{Bread}, {Jelly}, {PeanutButter}, {Bread,Jelly}, {Bread,PeanutButter}, {Jelly, PeanutButter}, {Bread,Jelly,PeanutButter}}
L2 ={{Bread}, {Milk}, {PeanutButter}, {Bread,Milk}, {Bread,PeanutButter}, {Milk, PeanutButter}, {Bread,Milk,PeanutButter}, {Beer}, {Beer,Bread}, {Beer,Milk}}
222CSE 5331/7331 F'2011
Partitioning Adv/Disadv
Advantages:– Adapts to available main memory– Easily parallelized– Maximum number of database scans is
two. Disadvantages:
– May have many candidates during second scan.
223CSE 5331/7331 F'2011
Parallelizing AR Algorithms
Based on Apriori Techniques differ:
– What is counted at each site– How data (transactions) are distributed
Data Parallelism– Data partitioned– Count Distribution Algorithm
Task Parallelism– Data and candidates partitioned– Data Distribution Algorithm
224CSE 5331/7331 F'2011
Count Distribution Algorithm(CDA)1. Place data partition at each site.2. In Parallel at each site do3. C1 = Itemsets of size one in I;4. Count C1;
5. Broadcast counts to all sites;6. Determine global large itemsets of size 1, L1;7. i = 1; 8. Repeat9. i = i + 1;10. Ci = Apriori-Gen(Li-1);11. Count Ci;
12. Broadcast counts to all sites;13. Determine global large itemsets of size i, Li;14. until no more large itemsets found;
226CSE 5331/7331 F'2011
Data Distribution Algorithm(DDA)1. Place data partition at each site.2. In Parallel at each site do3. Determine local candidates of size 1 to count;4. Broadcast local transactions to other sites;5. Count local candidates of size 1 on all data;6. Determine large itemsets of size 1 for local
candidates; 7. Broadcast large itemsets to all sites;8. Determine L1;9. i = 1; 10. Repeat11. i = i + 1;12. Ci = Apriori-Gen(Li-1);13. Determine local candidates of size i to count;14. Count, broadcast, and find Li;15. until no more large itemsets found;
228CSE 5331/7331 F'2011
Comparing AR Techniques Target Type Data Type Data Source Technique Itemset Strategy and Data Structure Transaction Strategy and Data Structure Optimization Architecture Parallelism Strategy
231CSE 5331/7331 F'2011
Incremental Association Rules Generate ARs in a dynamic database. Problem: algorithms assume static
database Objective:
– Know large itemsets for D– Find large itemsets for D {D D}
Must be large in either D or D D Save Li and counts
232CSE 5331/7331 F'2011
Note on ARs Many applications outside market
basket data analysis– Prediction (telecom switch failure)– Web usage mining
Many different types of association rules– Temporal– Spatial– Causal
233CSE 5331/7331 F'2011
Advanced AR Techniques
Generalized Association Rules Multiple-Level Association Rules Quantitative Association Rules Using multiple minimum supports Correlation Rules
234CSE 5331/7331 F'2011
Measuring Quality of Rules
Support Confidence Interest Conviction Chi Squared Test
235CSE 5331/7331 F'2011
Data Mining Outline
PART I - Introduction PART II – Core Topics
– Classification– Clustering– Association Rules
PART III – Related Topics
236CSE 5331/7331 F'2011
Related Topics Outline
Database/OLTP Systems Fuzzy Sets and Logic Information Retrieval(Web Search Engines) Dimensional Modeling Data Warehousing OLAP/DSS Statistics Machine Learning Pattern Matching
Goal: Examine some areas which are related to data mining.
237CSE 5331/7331 F'2011
DB & OLTP Systems Schema
– (ID,Name,Address,Salary,JobNo) Data Model
– ER– Relational
Transaction Query:
SELECT NameFROM TWHERE Salary > 100000
DM: Only imprecise queries
CSE 5331/7331 F'2011 238
Fuzzy Sets Outline
Introduction/Overview
Material for these slides obtained from:
Data Mining Introductory and Advanced Topics by Margaret H. Dunham
http://www.engr.smu.edu/~mhd/bookIntroduction to “Type-2 Fuzzy Logic” by Jenny Carter
http://www.cse.dmu.ac.uk/~jennyc/
239CSE 5331/7331 F'2011
Fuzzy Sets and Logic Fuzzy Set: Set membership function is a real valued
function with output in the range [0,1]. f(x): Probability x is in F. 1-f(x): Probability x is not in F. EX:
– T = {x | x is a person and x is tall}– Let f(x) be the probability that x is tall– Here f is the membership function
DM: Prediction and classification are fuzzy.
240CSE 5331/7331 F'2011
Fuzzy Sets and Logic
Fuzzy Set: Set membership function is a real valued function with output in the range [0,1].
f(x): Probability x is in F. 1-f(x): Probability x is not in F. EX:
– T = {x | x is a person and x is tall}– Let f(x) be the probability that x is tall– Here f is the membership function
243CSE 5331/7331 F'2011
Fuzzy Set Theory
A fuzzy subset A of U is characterized by a membership function
(A,u) : U [0,1]which associates with each element u of
U a number (u) in the interval [0,1] Definition
– Let A and B be two fuzzy subsets of U. Also, let ¬A be the complement of A. Then,» (¬A,u) = 1 - (A,u) » (AB,u) = max((A,u), (B,u))» (AB,u) = min((A,u), (B,u))
244CSE 5331/7331 F'2011
The world is imprecise. Mathematical and Statistical techniques often
unsatisfactory.– Experts make decisions with imprecise data in an
uncertain world.– They work with knowledge that is rarely defined
mathematically or algorithmically but uses vague terminology with words.
Fuzzy logic is able to use vagueness to achieve a precise answer. By considering shades of grey and all factors simultaneously, you get a better answer, one that is more suited to the situation.
© Jenny Carter
245CSE 5331/7331 F'2011
Fuzzy Logic then . . . is particularly good at handling uncertainty,
vagueness and imprecision. especially useful where a problem can be
described linguistically (using words). Applications include:
– robotics– washing machine control– nuclear reactors– focusing a camcorder– information retrieval– train scheduling
© Jenny Carter
247CSE 5331/7331 F'2011
Fuzzy Sets
The shape you see is known as the membership function
© Jenny Carter
248CSE 5331/7331 F'2011
Fuzzy Sets
Shows two membership functions: ‘tall’ and ‘short’
© Jenny Carter
249CSE 5331/7331 F'2011
NotationFor the member, x, of a discrete set with membership µ we use the notation µ/x . In other words, x is a member of the set to degree µ. Discrete sets are written as:
A = µ1/x1 + µ2/x2 + .......... + µn/xn
Or
where x1, x2....xn are members of the set A and µ1, µ2, ...., µn are their degrees of membership. A continuous fuzzy set A is written as:
© Jenny Carter
250CSE 5331/7331 F'2011
Fuzzy Sets The members of a fuzzy set are members to
some degree, known as a membership grade or degree of membership.
The membership grade is the degree of belonging to the fuzzy set. The larger the number (in [0,1]) the more the degree of belonging. (N.B. This is not a probability)
The translation from x to µA(x) is known as fuzzification.
A fuzzy set is either continuous or discrete. Graphical representation of membership
functions is very useful.
© Jenny Carter
251CSE 5331/7331 F'2011
Fuzzy Sets - Example
Again, notice the overlapping of the sets reflecting the real worldmore accurately than if we were using a traditional approach.
© Jenny Carter
252CSE 5331/7331 F'2011
Rules Rules often of the form:
IF x is A THEN y is B
where A and B are fuzzy sets defined on the universes of discourse X and Y respectively.
– if pressure is high then volume is small;– if a tomato is red then a tomato is ripe.
where high, small, red and ripe are fuzzy sets.
© Jenny Carter
CSE 5331/7331 F'2011 253
Information Retrieval Outline
Introduction/Overview
Material for these slides obtained from:Modern Information Retrieval by Ricardo Baeza-Yates and Berthier Ribeiro-Neto
http://www.sims.berkeley.edu/~hearst/irbook/Data Mining Introductory and Advanced Topics by Margaret H. Dunham
http://www.engr.smu.edu/~mhd/book
254CSE 5331/7331 F'2011
Information Retrieval
Information Retrieval (IR): retrieving desired information from textual data.
Library Science Digital Libraries Web Search Engines Traditionally keyword based Sample query:
Find all documents about “data mining”.
DM: Similarity measures; Mine text/Web data.
255CSE 5331/7331 F'2011
Information Retrieval
Information Retrieval (IR): retrieving desired information from textual data.
Library Science Digital Libraries Web Search Engines Traditionally keyword based Sample query:
Find all documents about “data mining”.
256CSE 5331/7331 F'2011
DB vs IR
Records (tuples) vs. documents Well defined results vs. fuzzy results DB grew out of files and traditional
business systesm IR grew out of library science and need
to categorize/group/access books/articles
257CSE 5331/7331 F'2011
DB vs IR (cont’d)
Data retrieval which docs contain a set of keywords? Well defined semantics a single erroneous object implies failure!
Information retrieval information about a subject or topic semantics is frequently loose small errors are tolerated
IR system: interpret contents of information items generate a ranking which reflects relevance notion of relevance is most important
258CSE 5331/7331 F'2011
Motivation
IR in the last 20 years: classification and categorization systems and languages user interfaces and visualization
Still, area was seen as of narrow interestAdvent of the Web changed this perception
once and for all universal repository of knowledge free (low cost) universal access no central editorial board many problems though: IR seen as key to finding the
solutions!
259CSE 5331/7331 F'2011
Basic Concepts
Logical view of the documents
Document representation viewed as a continuum: logical view of docs might shift
structure
Accentsspacing stopwords
Noungroups stemming
Manual indexingDocs
structure Full text Index terms
© Baeza-Yates and Ribeiro-Neto
260CSE 5331/7331 F'2011
UserInterface
Text Operations
Query Operations Indexing
Searching
Ranking
Index
Text
query
user need
user feedback
ranked docs
retrieved docs
logical viewlogical view
inverted file
DB Manager Module
Text Database
Text
The Retrieval Process
© Baeza-Yates and Ribeiro-Neto
261CSE 5331/7331 F'2011
Information Retrieval
Similarity: measure of how close a query is to a document.
Documents which are “close enough” are retrieved.
Metrics:– Precision = |Relevant and Retrieved|
|Retrieved|– Recall = |Relevant and Retrieved|
|Relevant|
262CSE 5331/7331 F'2011
Indexing
IR systems usually adopt index terms to process queries
Index term:– a keyword or group of selected words– any word (more general)
Stemming might be used:– connect: connecting, connection, connections
An inverted file is built for the chosen index terms
© Baeza-Yates and Ribeiro-Neto
263CSE 5331/7331 F'2011
Indexing Docs
Information Need
Index Terms
doc
query
Rankingmatch
© Baeza-Yates and Ribeiro-Neto
264CSE 5331/7331 F'2011
Inverted Files There are two main elements:
– vocabulary – set of unique terms – Occurrences – where those terms appear
The occurrences can be recorded as terms or byte offsets
Using term offset is good to retrieve concepts such as proximity, whereas byte offsets allow direct access
Vocabulary Occurrences (byte offset)… …
© Baeza-Yates and Ribeiro-Neto
265CSE 5331/7331 F'2011
Inverted Files
The number of indexed terms is often several orders of magnitude smaller when compared to the documents size (Mbs vs Gbs)
The space consumed by the occurrence list is not trivial. Each time the term appears it must be added to a list in the inverted file
That may lead to a quite considerable index overhead
© Baeza-Yates and Ribeiro-Neto
266CSE 5331/7331 F'2011
Example Text:
Inverted file
1 6 12 16 18 25 29 36 40 45 54 58 66 70
That house has a garden. The garden has many flowers. The flowers are beautiful
beautiful
flowers
garden
house
70
45, 58
18, 29
6
Vocabulary Occurrences
© Baeza-Yates and Ribeiro-Neto
267CSE 5331/7331 F'2011
Ranking
A ranking is an ordering of the documents retrieved that (hopefully) reflects the relevance of the documents to the query
A ranking is based on fundamental premisses regarding the notion of relevance, such as:– common sets of index terms– sharing of weighted terms– likelihood of relevance
Each set of premisses leads to a distinct IR model© Baeza-Yates and Ribeiro-Neto
268CSE 5331/7331 F'2011
Classic IR Models - Basic Concepts
Each document represented by a set of representative keywords or index terms
An index term is a document word useful for remembering the document main themes
Usually, index terms are nouns because nouns have meaning by themselves
However, search engines assume that all words are index terms (full text representation)
© Baeza-Yates and Ribeiro-Neto
269CSE 5331/7331 F'2011
Classic IR Models - Basic Concepts
The importance of the index terms is represented by weights associated to them
ki- an index term
dj - a document
wij - a weight associated with (ki,dj)
The weight wij quantifies the importance of the index term for describing the document contents
© Baeza-Yates and Ribeiro-Neto
270CSE 5331/7331 F'2011
Classic IR Models - Basic Concepts
– t is the total number of index terms– K = {k1, k2, …, kt} is the set of all index terms
– wij >= 0 is a weight associated with (ki,dj)
– wij = 0 indicates that term does not belong to doc
– dj= (w1j, w2j, …, wtj) is a weighted vector associated with the document dj
– gi(dj) = wij is a function which returns the weight associated with pair (ki,dj)
© Baeza-Yates and Ribeiro-Neto
271CSE 5331/7331 F'2011
The Boolean Model
Simple model based on set theory Queries specified as boolean expressions
– precise semantics and neat formalism Terms are either present or absent. Thus,
wij {0,1} Consider
– q = ka (kb kc)
– qdnf = (1,1,1) (1,1,0) (1,0,0)
– qcc= (1,1,0) is a conjunctive component
© Baeza-Yates and Ribeiro-Neto
272CSE 5331/7331 F'2011
The Vector Model
Use of binary weights is too limiting Non-binary weights provide consideration for
partial matches These term weights are used to compute a
degree of similarity between a query and each document
Ranked set of documents provides for better matching
© Baeza-Yates and Ribeiro-Neto
273CSE 5331/7331 F'2011
The Vector Model
wij > 0 whenever ki appears in dj
wiq >= 0 associated with the pair (ki,q)
dj = (w1j, w2j, ..., wtj)
q = (w1q, w2q, ..., wtq)
To each term ki is associated a unitary vector i The unitary vectors i and j are assumed to be
orthonormal (i.e., index terms are assumed to occur independently within the documents)
The t unitary vectors i form an orthonormal basis for a t-dimensional space where queries and documents are represented as weighted vectors
© Baeza-Yates and Ribeiro-Neto
274CSE 5331/7331 F'2011
Query Languages
Keyword Based Boolean Weighted Boolean Context Based (Phrasal & Proximity) Pattern Matching Structural Queries
© Baeza-Yates and Ribeiro-Neto
275CSE 5331/7331 F'2011
Keyword Based Queries
Basic Queries– Single word– Multiple words
Context Queries– Phrase– Proximity
© Baeza-Yates and Ribeiro-Neto
276CSE 5331/7331 F'2011
Boolean Queries
Keywords combined with Boolean operators:– OR: (e1 OR e2)
– AND: (e1 AND e2)
– BUT: (e1 BUT e2) Satisfy e1 but not e2
Negation only allowed using BUT to allow efficient use of inverted index by filtering another efficiently retrievable set.
Naïve users have trouble with Boolean logic.
© Baeza-Yates and Ribeiro-Neto
277CSE 5331/7331 F'2011
Boolean Retrieval with Inverted Indices
Primitive keyword: Retrieve containing documents using the inverted index.
OR: Recursively retrieve e1 and e2 and take union of results.
AND: Recursively retrieve e1 and e2 and take intersection of results.
BUT: Recursively retrieve e1 and e2 and take set difference of results.
© Baeza-Yates and Ribeiro-Neto
278CSE 5331/7331 F'2011
Phrasal Queries
Retrieve documents with a specific phrase (ordered list of contiguous words)– “information theory”
May allow intervening stop words and/or stemming.– “buy camera” matches:
“buy a camera” “buying the cameras” etc.
© Baeza-Yates and Ribeiro-Neto
279CSE 5331/7331 F'2011
Phrasal Retrieval with Inverted Indices
Must have an inverted index that also stores positions of each keyword in a document.
Retrieve documents and positions for each individual word, intersect documents, and then finally check for ordered contiguity of keyword positions.
Best to start contiguity check with the least common word in the phrase.
© Baeza-Yates and Ribeiro-Neto
280CSE 5331/7331 F'2011
Proximity Queries
List of words with specific maximal distance constraints between terms.
Example: “dogs” and “race” within 4 words match “…dogs will begin the race…”
May also perform stemming and/or not count stop words.
© Baeza-Yates and Ribeiro-Neto
281CSE 5331/7331 F'2011
Pattern Matching
Allow queries that match strings rather than word tokens.
Requires more sophisticated data structures and algorithms than inverted indices to retrieve efficiently.
© Baeza-Yates and Ribeiro-Neto
282CSE 5331/7331 F'2011
Simple Patterns
Prefixes: Pattern that matches start of word.– “anti” matches “antiquity”, “antibody”, etc.
Suffixes: Pattern that matches end of word:– “ix” matches “fix”, “matrix”, etc.
Substrings: Pattern that matches arbitrary subsequence of characters.– “rapt” matches “enrapture”, “velociraptor” etc.
Ranges: Pair of strings that matches any word lexicographically (alphabetically) between them.– “tin” to “tix” matches “tip”, “tire”, “title”, etc.
© Baeza-Yates and Ribeiro-Neto
284CSE 5331/7331 F'2011
Dimensional Modeling View data in a hierarchical manner more as
business executives might Useful in decision support systems and mining Dimension: collection of logically related
attributes; axis for modeling data. Facts: data stored Ex: Dimensions – products, locations, date
Facts – quantity, unit price
DM: May view data as dimensional.
285CSE 5331/7331 F'2011
Dimensional Modeling
View data in a hierarchical manner more as business executives might
Useful in decision support systems and mining Dimension: collection of logically related
attributes; axis for modeling data. Facts: data stored Ex: Dimensions – products, locations, date
Facts – quantity, unit price
287CSE 5331/7331 F'2011
Multidimensional Schemas
Star Schema shows facts and dimensions– Center of the star has facts shown in fact tables– Outside of the facts, each diemnsion is shown
separately in dimension tables– Access to fact table from dimension table via join
SELECT Quantity, PriceFROM Facts, LocationWhere (Facts.LocationID = Location.LocationID) and(Location.City = ‘Dallas’)
– View as relations, problem volume of data and indexing
292CSE 5331/7331 F'2011
OLAP Online Analytic Processing (OLAP): provides more
complex queries than OLTP. OnLine Transaction Processing (OLTP): traditional
database/transaction processing. Dimensional data; cube view Visualization of operations:
– Slice: examine sub-cube.– Dice: rotate cube to look at another dimension.– Roll Up/Drill Down
DM: May use OLAP queries.
293CSE 5331/7331 F'2011
OLAP Introduction
OLAP by Example
http://perso.orange.fr/bernard.lupin/english/index.htm What is OLAP?
http://www.olapreport.com/fasmi.htm
294CSE 5331/7331 F'2011
OLAP Online Analytic Processing (OLAP): provides more
complex queries than OLTP. OnLine Transaction Processing (OLTP): traditional
database/transaction processing. Dimensional data; cube view Support ad hoc querying Require analysis of data Can be thought of as an extension of some of the basic
aggregation functions available in SQL OLAP tools may be used in DSS systems Multidimentional view is fundamental
295CSE 5331/7331 F'2011
OLAP Implementations MOLAP (Multidimensional OLAP)
– Multidimential Database (MDD)– Specialized DBMS and software system capable of
supporting the multidimensional data directly– Data stored as an n-dimensional array (cube)– Indexes used to speed up processing
ROLAP (Relational OLAP)– Data stored in a relational database– ROLAP server (middleware) creates the
multidimensional view for the user– Less Complex; Less efficient
HOLAP (Hybrid OLAP)– Not updated frequently – MDD– Updated frequently - RDB
297CSE 5331/7331 F'2011
OLAP Operations
Simple query – single cell in the cube Slice – Look at a subcube to get more
specific information Dice – Rotate cube to look at another
dimension Roll Up – Dimension Reduction; Aggregation Drill Down Visualization: These operations allow the
OLAP users to actually “see” results of an operation.
299CSE 5331/7331 F'2011
Decision Support Systems Tools and computer systems that assist
management in decision making What if types of questions High level decisions Data warehouse – data which supports
DSS
300CSE 5331/7331 F'2011
Unified Dimensional Model
Microsoft Cube View SQL Server 2005
http://msdn2.microsoft.com/en-us/library/ms345143.aspx
http://cwebbbi.spaces.live.com/Blog/cns!1pi7ETChsJ1un_2s41jm9Iyg!325.entry MDX AS2005
http://msdn2.microsoft.com/en-us/library/aa216767(SQL.80).aspx
301CSE 5331/7331 F'2011
Data Warehousing
“Subject-oriented, integrated, time-variant, nonvolatile” William Inmon
Operational Data: Data used in day to day needs of company.
Informational Data: Supports other functions such as planning and forecasting.
Data mining tools often access data warehouses rather than operational data.
DM: May access data in warehouse.
302CSE 5331/7331 F'2011
Operational vs. Informational
Operational Data Data Warehouse
Application OLTP OLAP
Use Precise Queries Ad Hoc
Temporal Snapshot Historical
Modification Dynamic Static
Orientation Application Business
Data Operational Values Integrated
Size Gigabits TerabitsLevel Detailed Summarized
Access Often Less Often
Response Few Seconds Minutes
Data Schema Relational Star/Snowflake
303CSE 5331/7331 F'2011
Statistics Simple descriptive models Statistical inference: generalizing a model
created from a sample of the data to the entire dataset.
Exploratory Data Analysis: – Data can actually drive the creation of the
model– Opposite of traditional statistical view.
Data mining targeted to business user
DM: Many data mining methods come from statistical techniques.
304CSE 5331/7331 F'2011
Pattern Matching (Recognition)
Pattern Matching: finds occurrences of a predefined pattern in the data.
Applications include speech recognition, information retrieval, time series analysis.
DM: Type of classification.
305CSE 5331/7331 F'2011
Image Mining Outline
Image Mining – What is it? Feature Extraction Shape Detection Color Techniques Video Mining Facial Recognition Bioinformatics
306CSE 5331/7331 F'2011
The 2000 ozone hole over the antarctic seen by EPTOMS http://jwocky.gsfc.nasa.gov/multi/multi.html#hole
307CSE 5331/7331 F'2011
Image Mining – What is it? Image Retrieval Image Classification Image Clustering Video Mining Applications
– Bioinformatics– Geology/Earth Science– Security– …
308CSE 5331/7331 F'2011
Feature Extraction
Identify major components of image Color Texture Shape Spatial relationships Feature Extraction & Image Processing
http://users.ecs.soton.ac.uk/msn/book/ Feature Extraction Tutorial
http://facweb.cs.depaul.edu/research/vc/VC_Workshop/presentations/pdf/daniela_tutorial2.pdf
309CSE 5331/7331 F'2011
Shape Detection
Boundary/Edge Detection Time Series – Eamonn Keogh
http://www.engr.smu.edu/~mhd/8337sp07/shapes.ppt
310CSE 5331/7331 F'2011
Color Techniques
Color Representations
RGB:
http://en.wikipedia.org/wiki/Rgb
HSV: http://en.wikipedia.org/wiki/HSV_color_space
Color Histogram Color Anglogram
http://www.cs.sunysb.edu/~rzhao/publications/VideoDB.pdf
312CSE 5331/7331 F'2011
Video Mining
Boundaries between shots Movement between frames ANSES:
http://mmir.doc.ic.ac.uk/demos/anses.html
313CSE 5331/7331 F'2011
Facial Recognition Based upon features in face Convert face to a feature vector Less invasive than other biometric
techniques http://www.face-rec.org http://computer.howstuffworks.com/facial-re
cognition.htm SIMS:
http://www.casinoincidentreporting.com/Products.aspx
314CSE 5331/7331 F'2011
Microarray Data Analysis Each probe location associated with gene Measure the amount of mRNA Color indicates degree of gene expression Compare different samples (normal/disease) Track same sample over time Questions
– Which genes are related to this disease?– Which genes behave in a similar manner?– What is the function of a gene?
Clustering– Hierarchical– K-means
315CSE 5331/7331 F'2011
Affymetrix GeneChip® Array
http://www.affymetrix.com/corporate/outreach/lesson_plan/educator_resources.affx
316CSE 5331/7331 F'2011
Microarray Data - Clustering
"Gene expression profiling identifies clinically relevant subtypes of prostate cancer"
Proc. Natl. Acad. Sci. USA, Vol. 101, Issue 3, 811-816, January 20, 2004