Mining Object, Spatial,

Multimedia, Text, and Web Data

Data Mining
 Mining Complex Types of Data
 Mining spatial data
 Mining image data
 Mining text data
 Mining the Web
 Mining Spatial Databases
 Spatial database
 Space related data: maps, VLSI layouts, …
 Topological, distance information organized by spatial
indexing structures
 Spatial data warehousing
 Issue: different representations & structures
 Dimensions
 Nonspatial: 25-30 degree  hot
 Spatial-to-nonspatial: “New York”  “western provinces”
 Spatial-to-spatial: equi. temp region  0-5 degree region
 Measures
 numerical
 Spatial: collection of spatial pointers (0-5 degree region)
 Example: BC Weather Pattern
Analysis
 Input
 A map with about 3,000 weather probes scattered in B.C.
 Daily data for temperature, wind velocity, etc.
 Concept hierarchies for all attributes
 Output
 A map that reveals patterns: merged (similar) regions
 Goals
 Interactive analysis (drill-down, slice, dice, pivot, roll-up)
 Fast response time, Minimizing storage space used
 Challenge
 A merged region may contain hundreds of “primitive”
regions (polygons)
 Spatial Merge
 Precomputing: too much
storage space
 On-line merge: very
expensive
 Spatial Association Analysis
 Spatial association rule: A  B [s%, c%]
 A and B are sets of spatial or nonspatial predicates
 Topological relations: intersects, overlaps, disjoint, etc.
 Spatial orientations: left_of, west_of, under, etc.
 Distance information: close_to, within_distance, etc.
 Example
 is_a(x, “school”) ^ close_to(x, “sports_center”)
 close_to(x, “park”) [7%, 85%]
 Progressive Refinement
 First search for rough relationship (e.g. g_close_to for
close_to, touch, intersect) using rough evaluation (e.g.
MBR)
 Then apply only to those objects which have passed the
rough test
 Spatial Classification
 Spatial classification
 Analyze spatial objects to derive classification schemes,
such as decision trees in relevance to spatial properties
 Example
 Classify regions into rich vs. poor
 Properties: containing university, containing highway, near
ocean, etc.
 Spatial Cluster Analysis
 Constraints-based clustering
 Selection of relevant objects before clustering
 Parameters as constraints
 K-means, density-based: radius, min points
 Clustering with obstructed distance

C2 C3

r i dge
B C1
River

Mountain C4

Spatial data with obstacles Clustering without taking

obstacles into consideration
 Mining Image Data - Retrieval
 Description-based retrieval systems
 Retrieval based on image descriptions, such as keywords,
captions, size, etc.
 Labor-intensive, poor quality
 Content-based retrieval systems
 Retrieval based on the image content(features), such as
color histogram, texture, shape, and wavelet transforms
 Sample-based queries
 Find all of the images that are similar to the features of given
image
 Feature specification queries
 Specify or sketch image features like color, texture, or shape,
which are translated into a feature vector
Mining Image Data - Retrieval
Combining searches

Search for “blue sky” Search for “airplane in blue sky”

(top layout grid is blue) (top layout grid is blue and
keyword = “airplane”)
 Classification of Image Data
 Classification
 Decision tree
 Based on descriptive features
 Based on content features
 Feature extraction
 Extract features for classification from raw image
 Various image analysis techniques are required
 Data transformation, edge detection, etc.
 Example
 Classify sky images to recognize galaxies, stars, etc.
 By using properties obtained from image analysis
Classification of Image Data
 Mining Text Databases
 Text databases (document databases)
 Large collections of documents from various sources
 News articles, research papers, books, e-mail messages, and
Web pages
 Data stored is usually semi-structured
 Traditional information retrieval techniques become
inadequate for the increasingly vast amounts of text data
 Information retrieval
 Information is organized into documents
 Information retrieval problem
 Locating relevant documents based on user input, such as
keywords or example documents
 Basic Measures for IR
 Precision: the percentage of retrieved documents that are in
fact relevant to the query (i.e., “correct” responses)
| {Relevant}  {Retrieved} |
precision 
| {Retrieved} |
 Recall: the percentage of documents that are relevant to the
query and were, in fact, retrieved
| {Relevant}  {Retrieved } |
recall 
| {Relevant} |
 Keyword-Based Retrieval
 A document is represented by a set of keywords
 Retrieval by keyword matching
 Queries may use expressions of keywords
 (Car and accessory), (C++ or Java)
 Major difficulties
 Synonymy: same meaning but different word
 Ex> Q: “software”  Doc: about programming, do not have
the keyword
 Polysemy: same word but different meaning
 Ex> Q: “mining”  Doc: about gold mining, have the
keyword
 Similarity-Based Retrieval
 A document is represented as a keyword vector
 Retrieval by similarity computing
 Basic techniques
 Stop list – set of words that are frequent but irrelevant
 Ex> a, the, of, for, with, …
 Stemming – use a common word stem
 Ex> drug, drugs, drugged  drug
 Weighting – count frequency
 Term frequency, inverse document frequency, …
 Similarity metrics
 Measure the closeness of a document to a query
 Cosine similarity: v1  v2
sim(v1 , v2 ) 
| v1 || v2 |
 TF-IDF Weighting
 TF (Term Frequency)
 TF= f(t,d) : how many times term t appears in doc d
 More frequent  more relevant to topic
 Normalization:
 Document length varies : relative frequency preferred

 IDF (Inverse Document Frequency)

 IDF = 1 + log (n / k) : in how many documents term t appears
 n : total number of docs
 k : # docs with term t appearing (the document frequency)
 Less frequent among documents  more discriminative
 TF-IDF weighting
weight(t, d) = TF(t, d) * IDF(t)
 Latent Semantic Indexing
 Reduce the dimension of keyword matrix
 To resolve the synonym problem and the size problem
 Use a singular value decomposition (SVD) techniques
 Example
universe rocket moon car truck
D1  1 0 1 1 0 
D 2  0 1 1 0 0 

D3  1 0 0 0 0 
 
D4  0 0 0 1 1 
D5  0 0 0 1 0 
 
D6  0 0 0 0 1 
 SVD
 Singular Value Decomposition
 Decompose the matrix Amn
Amn = Umm Smn (Vnn)T
 Reduce dimension
 Select largest k singular values
A’mn = Umk Skk (Vnk)T
 Projection of A into k dimension
A’mn Vnk = Umk Skk
 Computing similarity
AAT = USVT(USVT)T
= USVTVSTUT
= (US)(US)T
 SVD
  0.75  0.29 0.28 0.00  0.53
  0.28  0.53  0.75 0.00   2.16 0.00 0.00 0.00 0.00
0 .29 0.00
 
 1.59 0.00 0.00 0.00
 0.20  0.19 0.45 0.58 0.63 
0.00 V  ...
T
U   S  0.00 0.00 1.28 0.00
  0 .45 0.63  0.20 0.00 0 .19   
  0.33 0.22 0.12  0.58 0.41  0.00 0.00 0.00 1.00 0.00
  0.00 0.00 0.00 0.00 0.39
 0.12 0.41  0.33 0.58  0.22

  0.62  0.46 1.00 0.78 0.40 0.47 0.74 0.10 

  0.60  0.84  1 .00 0. 88  0 .18 0. 16  0 . 54 
   
  0.04  0.30  1.00  0.62  0.32  0.87
AV  US 2    (US )(US )T   
 0 .97 1. 00  1 .00 0. 94 0. 93 
 
  0.71 0.35   1.00 0.74 
   
  0.26 0.65   1.00 
 Automatic Document
Classification
 Motivation
 Automatic classification for the tremendous number of on-line
text documents (Web pages, e-mails, etc.)
 A classification problem
 Training set: Human experts generate a training data set
 Classification(learning): The system discovers the
classification rules
 Methods
 Extract keywords and weights from documents
 Documents are represented as (keyword, weight) pairs
 Classify training documents into classes
 Apply classification algorithm
 Decision tree, Bayesian, neural network, etc.
 Mining the World-Wide Web
 WWW provides rich sources for data mining
 Contents information
 Hyperlink information
 Usage information
 Challenges
 Too huge for effective data warehousing and data mining
 Too complex and heterogeneous
 Growing and changing very rapidly
 Web Search Engines
 Index-based
 Search the Web, collect Web pages, index Web pages, and
build and store huge keyword-based indices
 Locate sets of Web pages containing certain keywords
 Deficiencies
 A topic of any breadth may easily contain hundreds of
thousands of documents
 Many documents that are highly relevant to a topic may not
contain keywords defining them (synonymy, polysemy)
 Web Contents Mining -
Classification
 Web page/site classification
 Assign a class label to each web page from a set of
predefined topic categories
 Based on a set of examples of preclassified documents
 Example
 Use Yahoo!'s taxonomy and its associated documents as
training and test sets
 Derive a Web document classification model
 Use the model to classify new Web documents by assigning
categories from the same taxonomy
 Methods
 Keyword-based classification, use of hyperlink information,
statistical models, …
 Web Structure Mining
 Finding authoritative Web pages
 Retrieving pages that are not only relevant, but also of high
quality, or authoritative on the topic
 Hyperlinks can infer the notion of authority
 A hyperlink pointing to another Web page, this can be
considered as the author's endorsement of the other page
 Problems
 Not every hyperlink represents an endorsement
 One authority will seldom point to its rival authority
 Authoritative pages are seldom particularly descriptive
 Hub
 Set of Web pages that provides collections of links to
authorities
 HITS (Hyperlink-Induced
Topic Search)
 Method
1. Use an index-based search engine to form the root set
2. Expand the root set into a base set
 Include all of the pages that the root-set pages link to, and all
of the pages that link to a page in the root set
3. Apply weight-propagation
 Determines numerical estimates of hub and authority
weights
4. Output a list of the pages
 Large hub weights, large authority weights for the given
search topic
 Systems based on the HITS algorithm
 Clever, Google
 Achieve better quality search results than AltaVista, Yahoo!
 Web Usage Mining
 Mining Web log records
 Discover user access patterns
 Typical Web log entry - URL requested, the IP address from
which the request originated, timestamp, etc.
 OLAP on the Weblog database
 Find the top N users, top N accessed Web pages, most
frequently accessed time periods, etc.
 Data mining on Weblog records
 Find association patterns, sequential patterns, and trends of
Web accessing
 Web Usage Mining
 Applications
 Target potential customers for electronic commerce
 Identify potential prime advertisement locations
 Enhance the quality and delivery of Internet information
services to the end user
 Improve Web server system performance
 Web caching, Web page prefetching, and Web page swapping